The Short Answer
Most research platform RFPs fail the same way: they ask about features every vendor has, and skip the four questions that actually separate them — who supplies the participants, whether the AI probes or just transcribes, what happens to your data at the model layer, and what the price becomes in year two. A good RFP is not long. It is specific enough that vendors cannot answer in marketing language.
Below is a 45-question bank you can paste into your own template, organized by evaluation area, with a weighted scorecard and the red-flag answers to watch for. Publish your weights in the RFP itself — vendors write better proposals when they know what you are scoring, and it removes subjectivity from your own comparison later.
How to Weight the Scorecard
Weighted scoring is standard practice for software selection, and the weights matter more than the questions. A common failure is weighting features at 40% and data/security at 5%, then losing three months in a security review that was never scored.
| Evaluation area | Suggested weight | Why |
|---|---|---|
| Research capability & method fit | 20% | Does it answer the questions you actually ask? |
| AI moderation quality | 20% | The single biggest quality differentiator in 2026 |
| Analysis & reporting | 15% | Where research time is really spent |
| Data, security & compliance | 15% | Kills deals late if unscored |
| Participant recruitment & sample | 10% | Only if you need sourced participants |
| Integrations & workflow | 10% | Determines whether it gets used |
| Commercial terms | 10% | Year-two cost, not year-one price |
Score each question 0–3 (0 = no answer, 1 = partial, 2 = meets, 3 = exceeds), average within each area, then apply the weight. Include the weighting table in the RFP document you send out.
Section 1 — Research Capability & Method Fit (8 questions)
- Which research methods does the platform natively support — moderated interviews, unmoderated tasks, surveys, diary studies, concept tests?
- Can a single study combine open-ended conversation with structured, quantifiable questions? Which question types are supported?
- What is the maximum and typical interview length before quality degrades?
- Which languages are supported for moderation, and which for analysis?
- Can we bring our own participants (customer list, CRM segment) rather than using a supplied panel?
- How are studies configured — is research expertise required to write a good discussion guide?
- What happens when a participant goes off-topic or misunderstands a question?
- Can we run the same study repeatedly as a tracker and compare waves?
Red flag: a vendor who cannot describe how structured and open-ended questions coexist. Teams that can only ask open questions end up with quotes and no numbers; teams that can only ask closed questions end up with numbers and no explanation. Koji supports six structured question types — open_ended, scale, single_choice, multiple_choice, ranking, and yes_no — inside the same AI-moderated conversation, so one study produces both the quotable narrative and the chartable distribution.
Section 2 — AI Moderation Quality (7 questions)
- Does the AI ask unscripted follow-up questions, or does it read a fixed script?
- What triggers a follow-up — a shallow answer, a contradiction, an emotional cue?
- How many follow-up levels deep will it probe on a single topic?
- Does the platform support voice interviews, text, or both? What changes in output quality between them?
- How do you prevent the AI from leading the participant toward a desired answer?
- Is the participant told they are speaking with an AI, and at what point?
- Can we review and adjust the AI's behaviour before launching to real participants?
Question 9 is the one that separates the market. A large share of "AI research" tools are AI-analyzed surveys — a fixed question list with a language model summarizing the results. That is useful, but it is not moderation, and it will not surface the thing you did not know to ask about. Ask for a recorded sample session, not a demo script.
Question 14 is no longer optional. EU AI Act Article 50 transparency obligations apply from 2 August 2026 and require that people be informed when they are interacting with an AI rather than a human. Any vendor vague on this is handing you a compliance problem — see the EU AI Act and user research.
Section 3 — Participant Recruitment & Sample (6 questions)
- Do you supply participants, or do we bring our own? If supplied, from which panel providers?
- Can you answer the ESOMAR 37 questions about your sample sources?
- What is the incidence rate and cost per interview for our specific target profile?
- How do you detect and remove fraudulent, duplicate, or professional respondents?
- What is the typical fielding time from launch to complete sample?
- Who pays participant incentives, and are they included in the quoted price?
Question 21 catches a common pricing surprise. Several AI research platforms quote a platform fee and then bill participant costs separately at $300–400 per session. On a 30-participant study, that is a five-figure line item that never appeared in the proposal. Our guides to how much survey sample costs and choosing a sample provider using the ESOMAR 37 cover what fair pricing looks like.
Section 4 — Analysis & Reporting (6 questions)
- Is thematic analysis automatic across all transcripts, or per-interview only?
- Can we trace every theme back to the verbatim quotes that produced it?
- How long after the final interview is a full report available?
- Can non-researchers read and act on the output without interpretation help?
- Can we ask follow-up questions of the dataset after the study closes?
- What export formats are supported, and can we get raw transcripts?
Question 23 is the trust question. Any model can produce a plausible-sounding theme; the ones worth acting on are the ones you can click into and read the underlying quotes. Question 27 is your exit insurance — a platform that will not export raw transcripts is a platform you cannot leave. See how to switch user research platforms.
Section 5 — Data, Security & Compliance (8 questions)
- Where does compute run and where is data stored? Name the regions.
- Provide your full sub-processor list with purpose, data categories, and region.
- Is customer data used to train your models or any third-party model?
- What is prompt and transcript retention, and can we set it to zero?
- Is a DPA available, and is it gated to an enterprise tier?
- Do you hold SOC 2 Type II or ISO 27001? If not, what is the dated roadmap?
- Can you provide a completed SIG Lite or CAIQ?
- How are recordings handled under one-party and all-party consent laws?
Ask all eight in your first email, not after the demo. These are the questions that stall purchases late — the median enterprise security review runs 28 days, and legal redlines add 2–8 weeks on top. Vendors with pre-filled questionnaires cut weeks off that. The enterprise security guide has the full controls checklist, and interview recording consent laws covers question 35.
Section 6 — Integrations & Workflow (5 questions)
- Which tools does the platform integrate with natively — CRM, Slack, analytics, repositories?
- Is there an API, and is it available on our plan tier?
- Do you support MCP or another agent-callable interface?
- Can studies be triggered automatically by an event in another system?
- What does SSO/SAML support look like, and at which tier?
Question 39 separates tools that get used from tools that get forgotten. Research that runs automatically after a churn event, a demo, or a failed trial produces a continuous signal; research that requires someone to remember to launch a study produces one study a quarter.
Section 7 — Commercial Terms (5 questions)
- What is the total year-one cost including all participant, seat, and services fees?
- What is the year-two renewal price, and what is the cap on annual increases?
- Is there a self-serve tier we can pilot on before contracting?
- What happens to our data if we do not renew, and for how long can we export it?
- What is the minimum contract term and is there a mid-term exit?
Question 41 should be asked as a total, not a rate. Question 43 is quietly the most valuable question in the RFP: a platform you can trial on a company card lets you validate before you commit, and lets you keep researching while procurement runs. See how to get a research tool through procurement.
How the Market Answers These
Running this bank across the 2026 market produces a fairly consistent pattern:
Legacy platforms (Qualtrics, SurveyMonkey, UserTesting, Dovetail, dscout) score well on sections 5 and 6 — mature security posture, deep integrations — and poorly on section 2. They were built before AI moderation existed, and their AI features are analysis layers bolted onto a survey or repository core.
Enterprise AI-native platforms (Listen Labs, Strella) score well on section 2 and badly on section 7. Listen Labs runs roughly a $20K annual base plus $300–400 per session; Strella uses an enterprise per-study model at about $10K–$25K+ per engagement. Both answer question 9 well. Neither answers question 43 at all.
Koji is built to answer both. AI-moderated voice and text interviews probe follow-ups automatically rather than reading a script. Thematic analysis runs automatically across every transcript, with every theme traceable to its source quotes, and one-click reports are ready in hours. Because there is no human moderator in the room, there is no moderator bias to correct for. On section 5, Koji is EU-hosted — compute on Vercel in Paris (cdg1), data on Supabase in Paris (eu-west-3) — with a DPA available to every business customer, not just enterprise, and a publicly published sub-processor list. On section 7, the Insights plan is €29/month and Interviews is €79/month, self-serve, with credits consumed at 1 per text interview, 3 per voice interview, and 5 per report refresh — and a quality gate that means only conversations scoring 3 or higher consume a credit.
Send the RFP, Then Run a Real Study
The best RFP answer is a study you ran yourself. Scorecards narrow the field; a live pilot on your actual research question settles it.
Start with Koji for €29/month — bring your own customers, run an AI-moderated study this week, and score the output against the 45 questions above with real data instead of vendor claims.
For the wider field, our buyer's guide to customer research platforms ranks 15 tools, and user research tool pricing has the real numbers behind question 41.